MS25:用于机器学习的材料科学专注的基准数据集 原子间潜力
Tristan Maxson1, Ademola Soyemi1, Xinglong Zhang2,3
1Department of Chemical and Biological Engineering, University of Alabama, Tuscaloosa, Alabama 35487, United States.
本研究介绍了MS25,一个用于机器学习原子间潜力 (MLIP) 的基准数据集. 相当的MLIP在复杂材料上表现出卓越的性能,突出了对物理可观测的明确验证的需要.
科学领域:
- 材料科学 材料科学 材料科学
- 计算化学计算化学
- 机器学习 机器学习
背景情况:
- 机器学习原子间潜力 (MLIP) 对于模拟材料行为至关重要.
- 评估MLIP需要各种各样的,与材料相关的基准数据集.
- 现有的基准可能无法完全捕捉无序或催化系统的复杂性.
研究的目的:
- 引入MS25,用于MLIP评估的综合基准数据集.
- 在各种材料系统中比较五个MLIP架构的性能.
- 评估MLIP预测对于超出简单能量和力误差的物理可观测的可靠性.
主要方法:
- 开发了MS25基准数据集,包括MgO表面,液态水,热质石,催化Pt,高合金 (HEAs) 和Zr-氧化物.
- 训练并测试了五种MLIP架构:MACE,NequIP,Allegro,MTP和Torch-ANI. 这些架构包括:
- 根据传统指标 (能量,力,应力) 和衍生的物理可观察值 (格子常数,体积,反应障碍) 评估MLIP.
主要成果:
- 大多数MLIP在简单系统上取得了可比的准确性,但在复杂/无序系统 (HEAs,Zr-O) 上,等价模型的表现超过非等价模型的1.5-2倍.
- 低能量/力误差不能保证准确的物理可观测值,强调需要明确的验证.
- 对于热带石模型,观察到跨框架可转移性的显著限制.
- HEA和Zr-O数据集被证明是具有挑战性的,有效地区分MLIP架构.
结论:
- 基准测试应该优先考虑了解MLIP故障模式,可转移性和对可观测物的影响,而不是边际准确性增长.
- 对于复杂的材料系统,建议使用等价MLIP.
- 对于更简单的问题,诸如计算成本,推断速度和软件集成等因素是关键的决策驱动因素.
更多相关视频
13:56Probe Type II Band Alignment in One-Dimensional Van Der Waals Heterostructures Using First-Principles Calculations
Published on: October 12, 2019
05:37Rapid in-silico Battery Electrolyte Electrochemical Reaction Generation using 3T-VASP Multi-Scale Energy Minimization
Published on: August 22, 2025
相关概念视频
Intermolecular Forces and Physical Properties
Predicting Molecular Geometry
Intermolecular vs Intramolecular Forces
Maxwell-Boltzmann Distribution: Problem Solving
This distribution function f(v) is defined by saying that the expected number N (v1,v2) of particles with speeds between v1 and v2 is given by
Intermolecular Forces in Solutions
When the strengths of the intermolecular forces of attraction between solute and solvent species in a solution are no different than those present in the separated components, the solution is formed with no accompanying energy change. Such a solution is called an ideal solution. A mixture of ideal gases (or gases such as helium and argon,...
Van der Waals Equation
First, the attractive forces between molecules, which are stronger at higher densities and reduce the pressure, are considered by adding to the pressure a term equal to the square of the molar density multiplied by a positive coefficient a. Second, the volume...
